Abstract

I will be talking about one of our recent papers [1] (and time permitting, some detail from our related work [2]). When explaining AI behavior to humans, how is the communicated information being comprehended by the human explainee, and does it match what the explanation attempted to communicate? When can we say that an explanation is explaining something? We can answer this by leveraging theory of mind literature about the folk concepts that humans use to understand behavior. We will establish a framework of social attribution by the human explainee, which describes the function of explanations: the concrete information that humans comprehend from them. Specifically, effective explanations should be coherent (communicate information which generalizes to other contrast cases), complete (communicating an explicit contrast case, objective causes, and subjective causes), and interactive (surfacing and resolving contradictions to the generalization property through iterations). We will also see that many XAI mechanisms can be mapped to folk concepts of behavior. This allows us to uncover their modes of failure that prevent current methods from explaining effectively, and what is necessary to enable coherent explanations.

[1] https://arxiv.org/abs/2201.11239

[2] https://arxiv.org/abs/2201.11569

Bio:

Alon is a 3rd year PhD student under supervision of Prof. Yoav Goldberg. His main research involves explainability in AI and NLP - in particular human-driven explainability – on formalizing human aspects of AI explanation, and bridging XAI technologies with how people perceive them.

Michael is inviting you to a scheduled Zoom meeting.

Topic: NLIP Seminar Time: May 6, 2022 12:00 PM London

Join Zoom Meeting https://cl-cam-ac-uk.zoom.us/j/94112888558?pwd=aGN2Skg2UFlnUkxWMmFuRjV6SCs0dz09

Meeting ID: 941 1288 8558 Passcode: 420834